Longitudinal Multivariate Psychology

Emilio Ferrer
Visit to download the full and correct content document:
https://textbookfull.com/product/longitudinal-multivariate-psychology-emilio-ferrer/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...

Fundamentals of Satellite Remote Sensing Emilio

Chuvieco

https://textbookfull.com/product/fundamentals-of-satellite-
remote-sensing-emilio-chuvieco/

Multivariate Data Analysis Joseph F. Hair

https://textbookfull.com/product/multivariate-data-analysis-
joseph-f-hair/

Practical multivariate analysis 6th Edition Afifi

https://textbookfull.com/product/practical-multivariate-
analysis-6th-edition-afifi/

Practical multivariate analysis 6th Edition Afifi

https://textbookfull.com/product/practical-multivariate-
analysis-6th-edition-afifi-2/
Problems in Quantum Mechanics with Solutions 2nd
Edition Emilio D'Emilio

https://textbookfull.com/product/problems-in-quantum-mechanics-
with-solutions-2nd-edition-emilio-demilio/

Actions and invariants of algebraic groups Second

Edition Ferrer Santos

https://textbookfull.com/product/actions-and-invariants-of-
algebraic-groups-second-edition-ferrer-santos/

Longitudinal Data Analysis Autoregressive Linear Mixed

Effects Models Ikuko Funatogawa

https://textbookfull.com/product/longitudinal-data-analysis-
autoregressive-linear-mixed-effects-models-ikuko-funatogawa/

Multivariate Analysis for the Behavioral Sciences Kimmo

Vehkalahti

https://textbookfull.com/product/multivariate-analysis-for-the-
behavioral-sciences-kimmo-vehkalahti/

Financial Markets Theory: Equilibrium, Efficiency and

Information 2nd Edition Emilio Barucci

https://textbookfull.com/product/financial-markets-theory-
equilibrium-efficiency-and-information-2nd-edition-emilio-
barucci/
LONGITUDINAL MULTIVARIATE
PSYCHOLOGY

This volume presents a collection of chapters focused on the study of multivariate

change. As people develop and change, multivariate measurement of that change
and analysis of those measures can illuminate the regularities in the trajecto-
ries of individual development, as well as time-dependent changes in population
averages. As longitudinal data have recently become much more prevalent in
psychology and the social sciences, models of change have become increasingly
important. This collection focuses on methodological, statistical, and modeling
aspects of multivariate change, as well as applications of longitudinal models to
the study of psychological processes.
The volume is divided into three major sections: Extension of latent change
models, Measurement and testing issues in longitudinal modeling, and Novel
applications of multivariate longitudinal methodology. It is intended for advanced
students and researchers interested in learning about state-of-the-art techniques
for longitudinal data analysis, as well as understanding the history and develop-
ment of such techniques.

Emilio Ferrer is Professor in the Department of Psychology at the University

of California, Davis, US, and a member of the Graduate Groups in Biostatistics,
Education, and Human Development. His research interests include methods to
examine multivariate change and developmental processes.
Steven M. Boker is Professor of Quantitative Psychology at the Univer-
sity of Virginia, US, Director of the Human Dynamics Lab, Speaker-Director
of the LIFE Academy, and Director of the Center for Dynamics of Healthy
Development. His research concerns dynamical systems modeling, interpersonal
communication, and methods for the study of development over the lifespan.
Kevin J. Grimm is Professor of Psychology at Arizona State University, US, and
head of the Quantitative Research Methods PhD program. His research interests
include longitudinal data analysis and data mining.
 Multivariate Applications Series

Sponsored by the Society of Multivariate Experimental Psychology, the goal of this series
is to apply statistical methods to significant social or behavioral issues, in such a way so as to
be accessible to a nontechnical-oriented readership (e.g., non-methodological researchers,
teachers, students, government personnel, practitioners, and other professionals). Applica-
tions from a variety of disciplines such as psychology, public health, sociology, education,
and business are welcome. Books can be single- or multiple-authored or edited volumes
that: (1) demonstrate the application of a variety of multivariate methods to a single, major
area of research; (2) describe a multivariate procedure or framework that could be applied
to a number of research areas; or (3) present a variety of perspectives on a topic of interest
to applied multivariate researchers.
Anyone wishing to submit a book proposal should send the following: (1) author/title;
(2) timeline including completion date; (3) brief overview of the book’s focus, including
table of contents and, ideally, a sample chapter (or chapters); (4) a brief description of
competing publications; and (5) targeted audiences.
For more information, please contact the series editor, Lisa Harlow, at Department of
Psychology, University of Rhode Island, 10 Chafee Road, Suite 8, Kingston, RI 02881-
0808; phone (401) 874-4242; fax (401) 874-5562; or e-mail LHarlow@uri.edu.

• Cognitive Assessment: An Introduction to the Rule Space Method, written by Kikumi

K. Tatsuoka (2009)
• Structural Equation Modeling with AMOS: Basic Concepts, Applications, and Programming,
Second Edition written by Barbara M. Byrne (2010)
• Handbook of Ethics in Quantitative Methodology, co-edited by Abigail T. Panter & Sonya
K. Sterba (2011)
• Longitudinal Data Analysis: A Practical Guide for Researchers in Aging, Health, and Social
Sciences, co-edited by Jason T. Newsom, Richard N. Jones, and Scott M. Hofer (2011)
• Structural Equation Modeling with MPlus: Basic Concepts, Applications, and Programming
written by Barbara M. Byrne (2012)
• Understanding the New Statistics: Effect Sizes, Confidence Intervals, and Meta-Analysis
written by Geoff Cumming (2012)
• Frontiers of Test Validity Theory: Measurement, Causation and Meaning written by Keith
A. Markus and Denny Borsboom (2013)
• The Essence of Multivariate Thinking: Basic Themes and Methods, Second Edition written
by Lisa L. Harlow (2014)
• Longitudinal Analysis: Modeling Within-Person Fluctuation and Change written by Lesa
Hoffman (2015)
• Handbook of Item Response Theory Modeling: Applications to Typical Performance Assessment
co-edited by Steven P. Reise & Dennis Revicki (2015)
• Longitudinal Structural Equation Modeling: A Comprehensive Introduction written by Jason
T. Newsom (2015)
• Higher-Order Growth Curves and Mixture Modeling with Mplus: A Practical Guide by
Kandauda A. S. Wickrama, Tae Kyoung Lee, Catherine Walker O’Neal, & Frederick
O. Lorenz (2016)
• What If There Were No Significance Tests?: Classic Edition by Lisa L. Harlow, Stanley
A. Mulaik, James H. Steiger (2016)
LONGITUDINAL
MULTIVARIATE
PSYCHOLOGY

Edited by Emilio Ferrer, Steven M. Boker,

and Kevin J. Grimm
First published 2019
by Routledge
711 Third Avenue, New York, NY 10017
and by Routledge
2 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN
Routledge is an imprint of the Taylor & Francis Group, an informa business
© 2019 Taylor & Francis
The right of Emilio Ferrer, Steven M. Boker, and Kevin J. Grimm to be identified as the authors
of the editorial material, and of the authors for their individual chapters, has been asserted in
accordance with sections 77 and 78 of the Copyright, Designs and Patents Act 1988.
All rights reserved. No part of this book may be reprinted or reproduced or utilized
in any form or by any electronic, mechanical, or other means, now known or
hereafter invented, including photocopying and recording, or in any information
storage or retrieval system, without permission in writing from the publishers.
Trademark notice: Product or corporate names may be trademarks or registered trademarks,
and are used only for identification and explanation without intent to infringe.
Library of Congress Cataloging-in-Publication Data
Names: Ferrer, Emilio, author. | Boker, Steven M., author. | Grimm, Kevin J., author.
Title: Longitudinal multivariate psychology / Emilio Ferrer, Steven M. Boker, Kevin J. Grimm.
Description: New York, NY : Routledge, 2019. | Series: Multivariate applications series |
Includes bibliographical references and index.
Identifiers: LCCN 2018014758 | ISBN 9781138064225 (hardback) |
ISBN 9781138064232 (pbk.)
Subjects: LCSH: Psychology–Statistical methods. |
Psychology–Research–Longitudinal studies. | Multivariate analysis.
Classification: LCC BF39 .F47 2019 | DDC 150.1/519535–dc23
LC record available at https://lccn.loc.gov/2018014758
ISBN: 978-1-138-06422-5 (hbk)
ISBN: 978-1-138-06423-2 (pbk)
ISBN: 978-1-315-16054-2 (ebk)
Typeset in Bembo
by Out of House Publishing
CONTENTS

Foreword viii
John R. Nesselroade
List of Contributors x

Introduction and Section Overview 1

Emilio Ferrer, Steven M. Boker, & Kevin J. Grimm

SECTION I
Extensions of Latent Change Score Models 7

1 Methodological Issues and Extensions to the Latent

Difference Score Framework 9
Linying Ji & Sy-Miin Chow

2 Discrete- and Semi-Continuous Time Latent Change Score

Models of Fluid Reasoning Development from Childhood
to Adolescence 38
Emilio Ferrer

3 Individually Varying Time Metrics in Latent Change

Score Models 61
Kevin J. Grimm & Ross Jacobucci
vi Contents

4 Latent Change Score Models with Curvilinear Constant Bases 80

Fumiaki Hamagami & John J. McArdle

5 Regularized Estimation of Multivariate Latent Change

Score Models 109
Ross Jacobucci & Kevin J. Grimm

6 The Reticular Action Model: A Remarkably Lasting

Achievement 126
Steven M. Boker

SECTION II
Measurement and Testing Issues in
Longitudinal Modeling 143

7 Small Sample Corrections to Model Fit Criteria for Latent

Change Score Models 145
Sarfaraz Serang

8 On Interindividual Differences in Intraindividual Changes 166

Lijuan Wang & Miao Yang

9 Sample Size and Measurement Occasion Planning for Latent

Change Score Models through Monte Carlo Simulation 189
Zhiyong Zhang & Haiyan Liu

10 Investigating the Performance of cart- and Random

Forest-Based Procedures for Dealing with Longitudinal
Dropout in Small Sample Designs under mnar Missing Data 212
Timothy Hayes

11 From Factors of Curves to Factors of Change 240

Ryne Estabrook

SECTION III
Novel Applications of Multivariate
Longitudinal Methodology 261

12 The Role of Interval Measurement in Developmental Studies 263

Ryan P. Bowles
 Contents vii

13 Growth Modeling using the Differential Form: Translations

from Study of Fish Growth 277
Nilam Ram & Xiao Yang

14 Modeling Change with Data Collected from Relatives 301

Michael C. Neale

15 Making the Cut: How a Quantitative Psychologist Changed

College Sports 315
Todd A. Petr & Thomas S. Paskus

16 A Successful Consultation “Team” Model Applying

Contemporary Advanced Statistics to Minority
Research Centers 323
Earl S. Hishinuma, Deborah A. Goebert, Naleen N. Andrade,
Jane M. M. Onoye, Jeanelle J. Sugimoto-Matsuda, Junji Takeshita,
& Stephanie T. Nishimura

Index 341
FOREWORD
John R. Nesselroade

I am deeply honored to write this foreword to Jack McArdle’s festschrift, primarily

for two reasons. First, Jack is truly deserving of the recognition this volume
affords. Second, in addition to his prolific scholarship, innovative science, and
highly valued collegiality, Jack has been an “old school” mentor to a large num-
ber of researchers now outstanding in their own right. By virtue of both their
association with Jack and their own contributions, these scholars have earned the
right to speak about Jack’s science and their own. And herein, by dint of their
thoughtful and lasting contributions, they do so emphatically. This volume will
stand for a long time as a marker for an outstanding scientist who forged a great
career while pursuing his ambitious goals for the field. These chapters make it
abundantly clear that the case for obtaining and analyzing longitudinal data that
has been so vigorously prosecuted by Jack McArdle remains in good hands for the
foreseeable future.
Through most of the 1970s, I served as Secretary-Treasurer of the Society
of Multivariate Experimental Psychology (SMEP). In that capacity, I received a
letter one day from a young graduate student at Hofstra University asking how he
could become a member of SMEP. I carefully explained to the young man that
SMEP was a small scientific society, membership in which could only be obtained
through a process of nomination and voting by the active members of the society.
I encouraged him to complete his PhD, get an academic job, and be productive
in multivariate behavioral research and some day he could aspire to membership
in SMEP. My response probably sounded platitudinous, but it was meant to be
helpful.
Over the next decade or so, I had several occasions to connect with that young
scientist who had hoped to become a SMEP member. It was clear that his research
career was progressing in a most impressive fashion. He wanted psychology to
 Foreword ix

become a more rigorous science and he was both innovative and enthusiastic in
the way he went about doing what he thought would make it happen. When he
had the opportunity to apply for a faculty position in the Psychology Department
at the University of Virginia, I was delighted to write a letter of recommendation
on his behalf. Parenthetically, he returned the favor 15 years later when I was
under consideration for a position at UVA. Over the years we have collaborated
on a large number of papers, chapters, and a book. As our collaboration grew, so
did our friendship. For decades now, I have been richly blessed to have him as
both dear friend and esteemed colleague.
As the reader has likely guessed by now, that young SMEP aspirant was Jack
McArdle. In the course of building his brilliant scientific career, Jack’s early hope
was realized precisely and he was indeed elected to membership in SMEP. In fact,
Jack has served SMEP as its president. Yes, Jack made it and made it big! His
enduring legacy, aimed at helping to make psychology “a quantitative, rational
science,” is attested to by the present outstanding volume. The editors and authors
have done a great mentor proud!
John R. Nesselroade
March, 2017
CONTRIBUTORS

Naleen N. Andrade, Department of Psychiatry, University of Hawai‘i at Mānoa,

USA
Steven M. Boker, Department of Psychology, The University of Virginia, USA

Ryan P. Bowles, Department of Human Development and Family Studies,

Michigan State University, USA
Sy-Miin Chow, Department of Human Development and Family Studies,
Pennsylvania State University, USA
Ryne Estabrook, Department of Medical Social Sciences, Northwestern Univer-
sity, USA
Emilio Ferrer, Department of Psychology, University of California, Davis, USA

Deborah A. Goebert, Department of Psychiatry, University of Hawai‘i at Mānoa,

USA
Kevin J. Grimm, Department of Psychology, Arizona State University, USA

Fumiaki Hamagami, Longitudinal Research Institutes, USA

Timothy Hayes, Department of Psychology, Florida International University,

USA
Earl S. Hishinuma, Department of Psychiatry, University of Hawai‘i at Mānoa,
USA
Ross Jacobucci, Department of Psychology, University of Notre Dame, USA
 Contributors xi

Linying Ji, Department of Human Development and Family Studies, Pennsylvania

State University, USA
Haiyan Liu, Department of Psychology, University of Notre Dame, USA

John J. McArdle, Longitudinal Research Institutes, USA

Michael C. Neale, Virginia Institute for Psychiatric and Behavioral Genetics,

Virginia Commonwealth University, USA
John R. Nesselroade, Department of Psychology, University of Virginia, USA

Stephanie T. Nishimura, Department of Psychiatry, University of Hawai‘i at

Mānoa, USA
Jane M. M. Onoye, Department of Psychiatry, University of Hawai‘i at Mānoa,
USA
Thomas S. Paskus, National Collegiate Athletic Association, USA

Todd A. Petr, National Collegiate Athletic Association, USA

Nilam Ram, Department of Human Development and Family Studies, Pennsyl-

vania State University, USA
Sarfaraz Serang, Department of Psychology, University of Southern California,
USA
Jeanelle J. Sugimoto-Matsuda, Department of Psychiatry, University of Hawai‘i
at Mānoa, USA
Junji Takeshita, Department of Psychiatry, University of Hawai‘i at Mānoa, USA

Lijuan Wang, Department of Psychology, University of Notre Dame, USA

Miao Yang, Department of Psychology, University of Notre Dame, USA

Xiao Yang, Department of Human Development and Family Studies, Pennsylva-

nia State University, USA
Zhiyong Zhang, Department of Psychology, University of Notre Dame, USA
INTRODUCTION AND
SECTION OVERVIEW
Emilio Ferrer, Steven M. Boker, & Kevin J. Grimm

Most phenomena in psychology and the behavioral and social sciences involve
multiple processes that unfold over time. As such, much of this research is focused
on questions related to how and why constructs change over time. Addressing these
questions requires models that can capture changes and interrelations amongst con-
structs, as well as potential mechanisms underlying such dynamics. This is not an
easy task, and is a pursuit that has kept quantitative methodologists on their toes
for many years. One of the most influential leaders in this quest for the last three
decades has been John J. (Jack) McArdle. McArdle’s work has opened the path to
models for studying longitudinal measurement, growth, and change, and set the
stage for novel approaches to examine multivariate interrelations and dynamics.
Inthisvolume,anumberofhisformerstudents,colleagues,andfriendspayatribute
to McArdle’s work, attesting to his creative thinking, his innovative methodological
advantages, and the lasting legacy of his work. In particular, the volume presents
a collection of chapters focused on the study of multivariate change. As people
develop and change, multivariate measurement of that change and analysis of those
measures can illuminate the regularities in the trajectories of individual development
as well as time dependent changes in population averages. As longitudinal data have
recently become much more prevalent in psychology and the social sciences, models
of change have become increasingly important to provide a deep understanding
of mechanisms underlying development. The chapters focus on methodological,
statistical, and modeling aspects of multivariate change, as well as applications of
longitudinal models to the study of psychological processes.
The volume is divided into three major sections, each of them dealing with a
set of methodological issues.
2 Introduction

Section I: Extensions of Latent Change Score Models

Latent Change Score (LCS) models are a class of structural equation models for
examining change and dynamics. They are based on latent variables that repre-
sent changes in a process and focus on such changes and their mechanisms over
time. LCS models constitute a general modeling approach that can accommodate
multiple specifications, including one or several variables, determinants of those
changes, multiple groups, factor structures with various measures, and unobserved
classes, among others. This framework to study multivariate change, developed by
McArdle, is the main theme of this section.
The chapter by Linying Ji and Sy-Miin Chow discusses two important exten-
sions of the LCS for bivariate processes. The first extension includes the use
of alternative specifications of the initial conditions in the modeled trajectory.
Findings from their simulation study indicate that misspecifying initial condi-
tions of means and covariance matrix in a bivariate LCS model does not have
a strong effect on other model parameters. This is the case when the specified ini-
tial conditions are wide enough to cover the true initial conditions distribution.
Moreover, the effect of the potential misspecification is reduced with the avail-
ability of more time points. The second extension of the LCS model discussed in
Ji and Chow’s chapter features the use of stochasticity in the form of process noise
and measurement error. Their simulation results indicate that, although most of
the model parameters can be recovered with few time points under correct model
specification with as few as three time points, process noise cannot be reliably
distinguished from measurement error. This issue is magnified when initial con-
ditions are misspecified. Ji and Chow implement all models using their own R
statistical package dynr.
Ferrer presents an implementation of LCS modeling to study the develop-
ment of fluid reasoning and the contribution of brain structure maturation to
such development across childhood and adolescence. He uses an LCS model
that includes a measurement structure, thus, defining fluid reasoning as a latent
factor. Using data from a cohort-sequential design, he examines whether the
fluid reasoning factor is equivalent across measurement occasions and age through
factorial invariance analyses. He then specifies an LCS model across measurement
occasions and, with the use of discrete age bins, across age in a semi-continuous
fashion. Results from this latter model indicate that white matter volume
and structural integrity from the entire brain at any given age are related to
improvements in fluid reasoning the following year, throughout childhood and
adolescence.
The chapter by Grimm and Jacobucci discusses the implementation of LCS
models using a nonlinear mixed-effects modeling framework, where participants
can be measured at unique points in time. LCS models are typically specified using
discrete measurement occasions, which only take on a few unique values. If inter-
ested in a continuous time metric such as age, researchers commonly discretize
the time scales, thus relying on the assumption that any time-lag is constant within
 Introduction 3

and across participants, when this is not the case. Grimm and Jacobucci propose
using PROC NLMIXED in SAS and JAGS to estimate LCS models. Unlike stan-
dard structural equation modeling software, these programs allow the specification
of any number of time points, thus giving researchers a more fine-grained metric.
Hamagami and McArdle’s chapter presents an extension of LCS models based
on additive curvilinear effects. These effects are specified in the model together
with dynamic feedback and coupling parameters. A typical LCS specification
involves an additive latent component that is constant over time, while allow-
ing variation across individuals. In their chapter, Hamagami and McArdle extend
this specification by including an additive constant as a nonlinear function form,
and allowing the value of this latent constant to be time-dependent. For this, they
use a quadratic functional form. By incorporating the additive curvilinear effect,
the new LCS model can be used to examine dynamics in data that follow a wide
variety of nonlinear curves.
The chapter by Jacobucci and Grimm proposes regularized estimation for
examining a range of multivariate LCS models. Typically, determining the most
plausible LCS model amongst several alternative specifications involves fitting each
of the models separately and evaluating the corresponding fit statistics. Jacobucci
and Grimm’s proposed approach is a simultaneous evaluation of models based on
regularization using both maximum likelihood and Bayesian estimation. In addi-
tion to simplifying model selection, this approach allows additional parameteriza-
tions of the model, and increases the propensity of selecting the optimal model.
Boker’s chapter traces the history of path analysis and structural equation
modeling leading up to the so-called Reticular Action Model (RAM), one of
McArdle’s most influential methodological achievements. The RAM notation
allowed specifying structural equation models using three matrices only, instead
of the multiple matrices involved in standard programs at the time. This model
was implemented in software packages that allowed fitting models in a simplified
and parsimonious manner, and is still at the core of some of the most advanced
SEM programs. Boker’s chapter describes how several important technical diffi-
culties were solved in order to arrive at the RAM model. The chapter concludes
by discussing the evolution of structural equation modeling over the past 40 years.

Section II: Measurement and Testing Issues in

Longitudinal Modeling
In this section, a collection of chapters addresses issues related to measurement
and testing in longitudinal models. These issues cover a wide variety of topics,
including statistical fit criteria, sample size, design of measurement occasions,
incomplete data, and alternative ways to model change in multivariate data.
The chapter by Serang examines the performance of corrections implemented
to fit criteria when using LCS models with small samples. Using a simulation
study, Serang shows that the commonly used corrections tend to reduce the test
4 Introduction

statistic too much, leading to type I error rates below the nominal level, as well
as to a decrease in power. With the use of empirical data, the chapter illustrates
how such effects can yield spurious and even conflicting conclusions when fitting
equivalent models to the same data.
Wang and Yang’s chapter investigates the effects of over-simplifying random
components on the estimation of fixed effects in linear growth curves. Wang and
Yang use derivation work to demonstrate that, when both the random intercept
and the random slope are omitted from the specification, the standard errors of
the fixed-effects intercept and slope estimates can be biased. The authors repli-
cate their derivation findings using empirical data. The chapter concludes with
recommendations for researchers interested in fitting growth curves.
The chapter by Zhang and Liu deals with important design features for studies
that involve LCS models. Zhang and Liu carry out simulation work to determine
optimal sample size and number of measurement occasions for both univariate and
bivariate LCS models. They propose a method that allows researchers to calculate
the power for estimating individual model parameters using their own R statistical
package RAMpath.
Hayes’s chapter reports analyses examining the performance of two strategies
for dealing with missing data in longitudinal studies. Hayes compares CART (clas-
sification and regression trees) and random forest analyses when such techniques
are applied to multiple imputation and when they are used to model the proba-
bility of dropout through the creation of inverse probability weights. These two
approaches are evaluated under incomplete data due to “missing not at random”
(MNAR) mechanisms with small sample sizes. Hayes’s findings include that, in
small samples, CART and weighting methods outperform multiple imputation,
indicating that these approaches can help to deal with missing data problems in
empirical research.
In his chapter, Estabrook proposes a longitudinal perspective for measurement
evaluation, suggesting an extension of a measurement model based on derivatives.
This proposed approach builds on McArdle’s factor of curves (FOCUS) model,
in which the intercepts and slopes corresponding to growth curves of multiple
observed variables (or items) repeated over time are factor analyzed, yielding a
factor of the intercepts and a factor of the slopes. Estabrook’s proposed approach
provides a useful framework to address issues related to measurement evaluation
over time. This framework facilitates the integration of within-person change and
intra-individual processes into behavioral research.

Section III: Novel Applications of Multivariate

Longitudinal Methodology
The last section of this volume includes chapters highlighting the many contri-
butions by McArdle to developmental methodology and multivariate longitudinal
modeling at large. The chapters in this section cover topics such as measurement
 Introduction 5

and validity, growth and dynamics, and provide examples of current research on
which McArdle has left his indelibly mark, such as behavioral genetics, research
in collegiate athletics, and studies of minorities.
The chapter by Bowles focuses on measurement in developmental longitudinal
modeling. Bowles demonstrates how longitudinal studies are sensitive to violations
of interval level measurement, arguing that such violations have important impli-
cations regarding validity. Bowles describes three approaches to provide evidence
of interval scaling: mechanistic modeling, latent variable modeling, and Rasch
measurement. After discussing the strengths and limitations of each approach, the
chapter concludes with recommendations for empirical research.
Ram and Yang continue the theme of developmental methodology. In their
chapter, they focus on models for examining dynamics in developmental stud-
ies, for which they use techniques implemented in fisheries management. At
the core of their proposal is the integration of inter-individual differences with
intra-individual change. For this, they argue for a switch from the integrated
mathematical forms commonly used in growth modeling to differential equations.
Such equations, Ram and Yang maintain, allow a richer theoretical description
of the model and provide a more cogent articulation of the actual processes
underlying the change.
The chapter by Neale traces important developments in the genetic studies
of growth and development. Neale emphasizes the contributions to this research
by McArdle, including early developments in the psychometric and biometric
models, and more recent work encompassing multivariate latent growth curves
for twin data. In addition to reviewing these methodological developments, the
chapter describes potential extensions and applications of such models.
Petr and Paskus’s chapter depicts McArdle’s work for the National Collegiate
Athletic Association (NCAA). They narrate how, over 25 years, McArdle’s work
set the stage for high-level research at the NCAA that focused on how schools
can predict and promote the academic and life successes of college athletes. Petr
and Paskus describe how this work has changed, not just the research conducted
at the NCAA, but the very landscape of college athletics.
The final chapter in the volume, by Hishinuma and colleagues, illustrates
another important contribution by McArdle to current research, namely his work
on multivariate statistics as applied to research on minorities. Hishinuma and col-
leagues describe McArdle’s interactions with the personnel at the University of
Hawai’i at Mānoa and emphasize the profound impact of such interactions, not
only on the team but on the research area at large. The chapter includes personal
notes depicting McArdle’s rare combination of skills that allows him to blend into
and become part of an organization’s culture. These contributions, Hishinuma
and colleagues tell us, have been invaluable and have set the foundation for all
future endeavors of the research team.
In sum, we are pleased to present Longitudinal Multivariate Psychology. As its
title indicates, this volume is intended to offer a window into state-of-the-art
6 Introduction

longitudinal methodology for studying multivariate processes. It is our desire that

the volume fulfills our goal of providing researchers with templates for ideas on
how to examine change and dynamics. It is also our hope that this collection of
chapters honors and does justice to the creative, plentiful, and influential work by
Jack McArdle.
Emilio Ferrer
Steven M. Boker
Kevin J. Grimm
SECTION I

Extensions of Latent
Change Score Models
1
METHODOLOGICAL ISSUES AND
EXTENSIONS TO THE LATENT
DIFFERENCE SCORE FRAMEWORK1
Linying Ji & Sy-Miin Chow

Few topics in the social and behavioral sciences are as fundamental and as riddled
with controversies as the study of change. From the questions of whether and
how intervention programs impact the targeted behaviors of interest, to studies
of lifespan developmental changes, the pursuit of change arises in a variety of
contexts and brings its own fair share of methodological challenges. Particularly
well-known in the history of studying change are the various disputes surrounding
the analysis of observed difference scores (see e.g., Bereiter, 1963; Cattell, 1966;
Cronbach & Furby, 1970; Harris, 1963; Lord, 1956, 1958; Nesselroade & Cable,
1974), such as the low reliability of observed difference scores, and difficulties in
interpreting results from factor analyzing such scores. One of the key contribu-
tions of McArdle to the fields of psychometrics and lifespan development resides
in his proposal to formalize modeling of change scores at the latent variable level.
Among the latent difference models proposed by McArdle and colleagues, is the
univariate dual change score model, which allows researchers to examine data
manifesting nonlinear (specifically, sigmoid-shaped) curves as latent difference
score (LDS) models with linear change functions (McArdle & Hamagami, 2001).
LDS models allow researchers to capture change at the latent level, which
remedies the low reliability of change scores related directly to the measurement
errors of the testing tools. In addition, formulating change hypotheses within a
latent variable modeling framework also expands the ways in which researchers
may examine the determinants or pathways of intraindividual change and com-
plex interdependencies among multiple change processes – some of the rationales
for conducting longitudinal research proposed by Baltes and Nesselroade (1979).
Such work builds on the extensive literature on dynamic models in the form of
differential or difference equations.
Differential equations are used to represent a system of variables that evolve
continuously over time, whereas difference equation model are designed for
10 Linying Ji & Sy-Miin Chow

processes whose changes can be captured meaningfully in discrete time units. In

his work with colleagues, students, and collaborators, McArdle presented a latent
variable framework that reformulates differential equations measured at discrete,
equal time intervals into the form of structural equation models with latent
change components (Collins & Horn, 1991; Ferrer, McArdle, Shaywitz, Hola-
han, Marchione, & Shaywitz, 2007; Humphreys, 1996; McArdle & Hamagami,
2001; McArdle, 2001; McArdle & Hamagami, 2003a; Williams & Zimmerman,
1996). The resultant, discretized (in terms of time) models of latent change are
referred to herein as latent difference equation models.
Many phenomena of change can be expressed mathematically as difference
equation models. In the last two decades, difference equation models have been
steadily gaining traction as a modeling tool in the social and behavioral sciences.
For instance, time series models with autoregressive and/or moving average rela-
tions (Hamilton, 1994) can be readily formulated as special cases of difference
equation models. Models with such elements have been broadly utilized, for
instance, within the contexts of cross-lag panel models (Rogosa, 1978), longitudi-
nal mediation models (Cole & Maxwell, 2003), and dynamic factor analysis mod-
els (Browne & Nesselroade, 2005; Chow, Nesselroade, Shifren, & McArdle, 2004).
Beyond time series models, Cooney and Troyer (1994) proposed a nonlinear
associative memory model in the form of a difference equation as a way to rep-
resent individual and age differences in performance on the Sternberg memory
task. Another difference equation model that has evidenced a far-reaching impact
on models in the social and behavioral sciences is the logistic model. May (1974)
evaluated the logistic equation, also known as the Verhulst equation in contin-
uous form (Kaplan & Glass, 1995; Van Geert, 1993), in a difference equation
framework. He demonstrated that the logistic model can lead to a rich repertoire
of behaviors – a stable fixed point, limit cycle, and even chaotic behavior – but
only in the case of difference equations (for a review see: Hale & Koçak, 1991;
Kaplan & Glass, 1995). The logistic model has been used to capture changes in
ecological systems for decades. For example, Van Geert (1998) applied the discrete
logistic model and different variations of it to the context of language develop-
ment. In Newtson’s (1993) call for a greater synthesis between dynamical systems
concepts and research in mainstream social psychology, he also used the logistic
model as a proxy for representing human action and dyadic interactions.
In continuous form, the Verhulst equation has a known solution and only
leads to stable, fixed point behavior: that is, the process would always approach
an asymptote level (an equilibrium) exponentially and subsequently settle into
this equilibrium. The solution of the Verhulst equation yields a trajectory that is
similar in shape to other learning functions more familiar to psychologists, such
as the Gompertz curve. The Gompertz curve and other similar nonlinear growth
curve variations have been broadly studied and applied as plausible representations
of dynamic processes (Ram & Grimm, 2015). However, when nonlinearities are
present in the model, special linearization or other related procedures would have
 Extensions to LDS Framework 11

to be incorporated into the estimation routines to handle such nonlinearities (see

e.g., Browne & du Toit, 1991; Collins & Horn, 1991; Cudeck, 2002).
The dual change score (DCS) model is one of the latent difference equation
models proposed by McArdle and colleagues that could generate nonlinear curves
similar in form to the Gompertz curve — but as a linear difference equation
model without the need to utilize additional linearization procedures (McArdle,
2001; McArdle & Hamagami, 2001). Doing so opens up new possibilities for
representing growth and decline processes at the latent level without incurring
excessive computational burden. The bivariate dual change score (BDCS) model
is a bivariate extension of the dual change score model designed to represent
two parallel growth processes. Methodological evaluations and extensions of the
BDCS model have been considered by multiple researchers (Bollen & Curran,
2006; Duncan, Duncan, & Strycker, 2013; Ferrer, McArdle, Shaywitz, Holahan,
Marchione, & Shaywitz, 2007; McArdle, 2009; Singer & Willett, 2003). The
BDCS has also been used to represent myriad empirical constructs, including
the growth and decline of fluid and crystallized intelligence across the life-span
(McArdle, Hamagami, Meredith, & Bradway, 2000), the relation between anti-
social behaviors and reading achievement in adolescence (McArdle & Hamagami,
2001a),and the biometric genetic process in twin studies (McArdle & Hamagami,
2003b), among other applications.
In this chapter, we discuss two extensions to the BDCS model, including:
(1) alternative specifications of the initial conditions (ICs) in the BDCS model,
and how different IC specifications affect parameter estimation and interpreta-
tions of the change processes under study; and (2) the feasibility of incorporating
stochasticity at both the latent change as well as the measurement level. We refer
to stochasticity in the latent changes as stochasticity due to process noises. After
that, we evaluate the performance of the BDCS model when the model has
both misspecified IC and stochasticity in the latent changes. We demonstrate
the implementation of the stochastic BDCS model with the use of different ICs
using the longitudinal data of reading and arithmetic performance scores from
the Early Childhood Longitudinal Study, Kindergarten Class of 1998 (ECLS-K)
study. Estimation results from different variations of the stochastic BDCS model
are compared and the substantive implications are discussed. All model fitting
procedures in this chapter were performed using an R package, dynr (abbreviated
from Dynamic Modeling in R; Ou et al., 2017), for fitting linear and nonlinear
dynamic models with linear, Gaussian measurement models. R scripts for model
fitting can be found in the Appendix.

The Bivariate Dual Change Score (BDCS) Model

Let y1,it and y2,it be two observed variables for person i (i = 1, . . . , N) at time
t (t = 1, . . . T), linked to their associated latent true scores, denoted as η1,it and
η2,it , through a time-invariant measurement model. Let yit = [y1,it y2,it ] and
12 Linying Ji & Sy-Miin Chow

ηit = [η1,it η2,it ] . This measurement model can be expressed as

 
yit = ηit +  it ,  it ∼ N 0, diag[ω11 , ω22 ] (1.1)
where  it is a 2 × 1 vector of measurement errors for the two observed variables.
The BDCS model, just like many other longitudinal and dynamic models
whose components at any particular time point show dependencies on previous
components from one or more earlier time points, requires the specification of
the IC of the latent variables at time 1. The IC specification in the conventional
BDCS model will be discussed shortly. For now, we assume that beginning from
time t = 2, (for t = 2, . . ., T), the true scores for person i at time t are a function
of the person’s true scores at time t-1 plus latent changes as:
ηit = ηi,t−1 + ηit , (1.2)
The latent change vector, ηit , in turn, is assumed to conform to a dynamic
model as:
ηit = ηs,i + βηi,t−1 , where (1.3)
 
β1 γ1
β= ,
γ2 β2
in which β1 and β2 represent the self-feedback effect of the values of η1,i and η2,i
at the previous time point on their values at the current time point respectively;
γ1 and γ2 are the coupling parameters, representing the effects of η1,it−1 on η2,it
and of η2,it−1 on η1,it , respectively. ηs,i = [ηs1,i ηs2,i ] is a vector of constant (i.e.,
time-invariant) slopes, or additive components of change, that capture person
i’s constant amounts of change in the two variables of interest from one time
point to the next. If β is equal to a null matrix, then Equation 1.3 dictates that
the two processes show linear change trajectories throughout with constant but
person-specific amounts of change, captured by ηs1,i and ηs2,i , in successive time
points.
The dynamic model for ηit can also be written in a one-step-ahead form as
(Chow, Ho, Hamaker, & Dolan, 2010):
⎡ ⎤ ⎡ ⎤⎡ ⎤
η1,it 1 + β1 γ1 1 0 η1,it−1
⎢ η2,it ⎥ ⎢ γ2 1 + β2 0 1⎥ ⎢ ⎥
⎢ ⎥=⎢ ⎥ ⎢ η2,it−1 ⎥ , (1.4)
⎣ηs1,it ⎦ ⎣ 0 0 1 0⎦ ⎣ηs1,it−1 ⎦
ηs2,it 0 0 0 1 ηs2,it−1
where ηs1,it and ηs2,it are the constant slopes of the two processes at time t, now
explicitly constrained to be invariant over time to reflect the person-specific but
time-invariant nature of ηs,i . In the conventional BDCS model, the values of ηs1,it
and ηs2,it are constrained to be invariant over time. Incorporating these elements as

part of a latent variable vector, η∗it = η1,it η2,it ηs1,it ηs2,it as in Equation (1.4)
opens up the possibility for these constant slopes to vary over time. Equation 1.4
 Extensions to LDS Framework 13

may be regarded as a special case of the state-space model, or a vector autoregres-

sive model with person-specific intercepts. Here, 1 + β1 and 1 + β2 are equivalent
to what are traditionally referred to as the lag-1 autoregression parameters in the
time series literature.

IC Specifications in the BDCS

Since the BDCS model is a longitudinal model involving lagged dependencies
among successive measurement occasions, the specification of IC, namely, the
statistical properties of the data at the first available time point, is an important
issue. With assumptions of Gaussian distributed measurement processes, the IC of
the BDCS essentially includes specification of the mean and covariance structures
of the expanded latent variable vector, η∗i1 , at the first available observed occasion,
t = 1. At t = 1, McArdle and colleagues’ (2001b; 2001) original BDCS model
specifies that the initial structure of η∗i1 , takes on the form of:
η∗i1 = μ1 + ζ i , ζ i ∼ N(0, 1 ), with
 
μ1 = μη1 μη2 μηs1 μηs2 , ζ i = ζη1 ,i ζη2 ,i ζηs1 ,i ζηs2 ,i , and
⎡ ⎤
ψ11
⎢ψ21 ψ22 ⎥
⎢
η∗1 = ⎣
⎥, (1.5)
ψ31 ψ32 ψ33 ⎦
ψ41 ψ42 ψ43 ψ44

where η∗i1 = η1,i1 η2,i1 ηs1,i1 ηs2,i1 , in which η1,i1 and η2,i1 are the initial
levels of the two latent processes of interest at time 1; whereas ηs1,i1 and ηs2,i1 are
constant slopes for the two processes. Note that at least two time points are needed
to define a slope so it may be counterintuitive to impose an initial condition for
the slopes at the first time point. However, because of Equation 1.4, which dictates
the invariance of these constant slopes over time, these constant latent slopes are
defined by the changes across all time points, as opposed to information from
just the first time point. This specific form of IC, with all unique elements of
the mean vector and covariance matrix specified as freely estimated parameters, is
similar to the IC specification used in standard growth curve models (McArdle &
Epstein, 1987; Ram & Grimm, 2015). This IC specification coincides with the
so-called “free-parameter” approach considered by du Toit and Browne (2007)
in fitting SEM-based time series models and by other researchers in fitting state-
space models (Harvey & Souza, 1987). As in the original BDCS model, the free-
parameter approach also has the characteristic that all unknown parameters in
the initial condition distribution of the latent variables are specified to be freely
estimated parameters and as such, this approach allows the change processes under
study to be non-stationary (du Toit & Browne, 2007).
In addition to the “free-parameter” approach, another IC specification pro-
posed in the state-space framework, and further evaluated within the SEM
14 Linying Ji & Sy-Miin Chow

framework by several researchers (du Toit & Browne, 2007; Hamaker, 2005), is the
“model-implied” approach for partially stable processes. This approach assumes
stability2 in the portion of a modeled process that shows lagged dependencies,
which, in turn, requires all associated modeling parameters to be time invari-
ant. Imposing such assumptions then enables researchers to use the unconditional
mean and covariance matrix implied by the model to specify the initial distribu-
tion on the latent variables at the first observed time point. du Toit and Browne
(2007) showed that if the true underlying processes do in fact conform to a par-
tially stable structure and are well described by the “model implied” approach,
the IC structure can be recovered well using the “free-parameter” approach. We
thus do not discuss the “model-implied” approach further here.
Another IC specification commonly adopted in the state-space literature is a
diffuse IC specification in which the initial mean vector of the latent variables, μ1 ,
is set to a vector of zeros; and the corresponding initial latent variable covariance
matrix, 1 , is specified to be a diagonal matrix with variance entries fixed at arbi-
trarily large values (e.g., De Jong, 1991; Harvey, 2001; Harvey & Phillips, 1979;
Schweppe, 1973). Larger values of variances reflect greater uncertainties in the
structure of the IC. This IC specification assumes that limited information is avail-
able to specify the IC structure with high precision. Thus, the means of the initial
latent variables are all set to zero – a predicted value that is relatively reasonable
for many stable, time-intensive longitudinal processes (i.e., time series processes)
frequently considered in the econometric literature in which the diffuse IC speci-
fication is adopted. With the large numbers of time points typically available from
those studies, such misspecification of IC has relatively trivial consequences. This
is not the case, however, when panel data with a limited number of occasions is
involved (Chow et al., 2016; Losardo, 2012; Oud et al., 1990; Ou et al., 2016).
Indeed, the issue of IC specification has received considerable attention in the
state-space framework (De Jong, 1991; Harvey, 2001) and much can be learned
from some of the issues, problems, and solutions proposed in that literature. Pre-
vious studies have investigated the consequences of misspecification of IC in the
context of a vector autoregressive process with moving average residuals (du Toit
& Browne, 2007), an autoregressive latent trajectory model (Ou, Chow, Ji, &
Molenaar, 2016),and with both stable and unstable processes (Losardo, 2012).
However, the effects of IC misspecification have not been evaluated within the
context of the BDCS model. Our emphasis in this chapter is to evaluate whether
and to what extent misspecification in IC may interact with features of the differ-
ence equations implicated in the BDCS model. In particular, we examine whether
the BDCS model can reliably distinguish among variations due to interindividual
differences in IC, measurement errors, and process noises when model estimation
is performed using panel data with limited information.
As an initial demonstration of the consequences of IC misspecification in
the BDCS model in the absence of process noise, we simulated a set of data
using the BDCS model in Equations 1.1, 1.4, and 1.5; with N = 500 and two
configurations of total number of time points (T = 5 and T = 10). We then fit the
 Extensions to LDS Framework 15

BDCS model with the correctly specified IC structure, and the diffuse IC speci-
fication with all variance terms in 1 set to a sequence of possible values: .01, 1,
10, and 100. Table 1.1 summarizes the true values of the parameters used for data
generation purposes, as well as key simulation results based on fitting variations of
the BDCS model with freely estimated IC (the correctly specified IC), and mis-
specified IC specifications with different levels of “diffuseness.” We also show in
Table 1.1: (1) the total bias of the parameter estimates (over nine possible param-
eters), defined as the sum of the difference between each parameter’s estimated
and true values across all modeling parameters; and (2) the root mean squared
error (RMSE), defined as the square root of the mean of the squared differences
between the true and estimated values across all modeling parameters. In addition,
we also report the Akaiki Information Criterion (AIC) and Bayesian Information
Criteria (BIC) as model selection measures. We found that differences among the
conditions where the variances in 1 were set to 1, 10, and 100 were minimal.
Thus, we only included results from the condition with IC variances fixed at 10
due to space constraints.
As shown in Table 1.1, with T = 5 and the correctly specified IC, all of the
model parameters were recovered very well, with a small total bias and RMSE.
Both the AIC and the BIC selected the correctly specified IC structure over the
other IC structures. Of the conditions with diffuse IC specification, when the IC
covariance matrix was set to 1 = 10×I4 , only a slight increase in total bias was
observed. However, there was a five-fold increase in RMSE. When the number of
time points was increased from 5 to 10, all parameter estimates were very close to
their true values. The total bias and RMSE from the diffuse IC with 1 = 10 × I4
and T = 10 were close to those observed in the correctly specified IC. Also, as
expected, with larger T, smaller values of standard error estimates were observed.
The level of diffuseness of the IC also affected the quality of the point estimates.
In general, if the variances of the IC were specified to be large enough for the
misspecified IC distribution to show a reasonable amount of overlap with the true
IC distribution, estimation results were minimally affected. On the contrary, if the
variances of the IC were specified to be too restrictive – for instance, when some
of the interindividual variance terms were incorrectly fixed at 0, considerable
decrements in point estimates are noted. As an example, when we decreased the
IC variances to .01, all parameter estimates showed substantial biases and RMSEs,
even with 10 measurement occasions (see Table 1.1). The AIC and the BIC,
again, preferred models with more diffuse (e.g., 1 = 10 × I4 ) than the model
with the least diffuse but misspecified IC (1 = 0.01 × I4 ).
It should be noted that a sheer increase in the number of time points, per
se, may not always be adequate in overcoming issues pertaining to the misspec-
ification of the IC. For instance, when the absolute values of the self-feedback
parameters (i.e., the βs) are large, the changes in the latent processes (differences
in the latent processes between successive time points) would be larger and the
processes would reach their asymptotic levels faster. As shown in Figure 1.1, when
β1 = −.3 (the two plots in the top panel), the latent processes do not reach their
Another random document with
no related content on Scribd:
The Project Gutenberg eBook of
Menschenwee
This ebook is for the use of anyone anywhere in the United
States and most other parts of the world at no cost and with
almost no restrictions whatsoever. You may copy it, give it away
or re-use it under the terms of the Project Gutenberg License
included with this ebook or online at www.gutenberg.org. If you
are not located in the United States, you will have to check the
laws of the country where you are located before using this
eBook.

Title: Menschenwee
Roman van het land

Author: Is. Querido

Release date: September 30, 2023 [eBook #71767]

Language: Dutch

Original publication: Haarlem: De Erven F. Bohn, 1903

Credits: Jeroen Hellingman and the Online Distributed

Proofreading Team at https://www.pgdp.net/ for Project
Gutenberg

*** START OF THE PROJECT GUTENBERG EBOOK

MENSCHENWEE ***
[Inhoud]
[Inhoud]

MENSCHENWEE

[Inhoud]
MENSCHENWEE

ROMAN VAN HET LAND

DOOR
IS. QUÉRIDO

HAARLEM
DE ERVEN F. BOHN
Eerste Boek
 WINTER.

[1]
[Inhoud]
 EERSTE HOOFDSTUK.

[Inhoud]

I.

Diep, in zandkuil van uitgespitte hei, weggeslobberd met zijn

beenen, in blauw-zwarte glimmige klei, schoot ie rhytmisch-breed
over, zwaar-ademend, grauwe zandbonken van z’n stompkorte graaf
opwerpend in kracht-zwaai, àl meer roestige brokken en plakken
naar den omgespitten grond, die dichtgegooid lag boven hem.
Grauw-grijze december-donkering dreigde; gier-felle wind, met er
doorheen dwars gerafel van regen, bolde en loeide wijd-wild over het
verre middag-land, dat vaal-bruin, en groezelig-geel, aan de kim
vernevelde in toonloos grauw van winter-droef-doodsche akkers,
grimmig verdoffend tegen verschimd-bleeke bosschages en hagen
van ver-affe tuinderijen en bollenland. En zwaar-eindloos, laag,
stond boven het akkerbrons gestolpt de wild-dreigende lucht-
donkering, aanhollend wolkengrauw met al duisterder tintvegen. Hier
en daar brokkelde gloedloos pluimgroen van boerenkool en prei,
groen-vaal gerijd tusschen bollenakkers, die in vuil-geel dekriet
schimmelden of bemorst lagen met boomstronken en zwaren
takkenrommel. Flauwe silhouetten van grondwerkers en spitters
kontoerden in wrongig-gebuk, schonkig-somber, in den woest-kalen
omtrek van het grondbrok dat Dirk bewerkte. Zwaar-melancholisch
loomden van ver hun lijfbewegingen en starre armenheffing in het
stugge druillicht, dat àl droeviger door-donkerde over de velden.
Schonkig-gespierd, met zweet-vet op roodbruin gezicht, schoot Dirk
over, al hooger boven ’m, al dieper [2]wegzakkend in hei, uitspittend
onder z’n voeten in breeëre sneden, òpwerpend de zandvrachten,
vlijmend met scherpen kant van mes-blinkige spa in den grond, plots
harder en inspannender soms, met bloed-rood hoofd van
aanzwellende kracht-dreiging inhakkend en kervend, waar
kleibonken, ommodderend z’n beenen, kleef-zwaarder naar den
grond terugzogen.

Z’n vuile slijk-klompen klodderden in het sappige kleiblauw, en àl

zwaarder, werkkoortsig-heftig, in onberusten maatgang kerfde en
hakte z’n graaf door modder en zandgrond, in wilder opwerping van
àl zwarter gestikte brokken en roest, omwoelde en groef ie uit de
spitkuil, al breeër en dieper, tot het zwart-vette water uit den grond
borrel-schuimde rond z’n beenen en broek. Dan ging er ’n wellust-
tinteling door z’n werklijf, door z’n blaasbalgende borst, want dàt
moest ie zien van zijn land, doorwatering, vettig en modderig.

De ouë Gerrit Hassel, bijgenaamd de Blommepot, Dirks vader, was

achter z’n erf uit, de akkers opgeloopen, stond nou bibberend-
verkleumd voor de spithei op Dirk neer te kijken. Z’n lang, lichtelijk bij
schouders gekromd lijf, stramde ingebogen tegen den stoei-wind en
z’n rond gezicht vertrok, blauwig-koud, in wreveltrekken, voor zoover
’t kon wrevelen op z’n boersch-poppig gezicht, bol-rond, omzilverd
van grijs-prachtige lokken, flap-krullend en wind-stoeiend nu onder
z’n vuil-groen petje uit, met z’n zware sprookjes-mooie zilveren
baard, eerwaardig-bisschoppelijk vreemd omhuivend z’n
kindergezicht. Op z’n bruin-kalig verschoten jekker, spatte de regen
korrelig en z’n blauwe kiel wapperde ’n endje er onder uit, flapperig
lawaaiend. Toch had ie schik, de ouë Gerrit, want telkens berekende
ie met ’n stil-trotsch genot de zandvracht van elken schep, dien Dirk
in kracht-zwaai, breed-gebarend òp-wierp uit de kuil, boven ’m uit;
voelde ie zich overmoedig op dat vijfsteek-diep-spitten van z’n
jongen, den eenigen van wien ie ’n beetje hield, om z’n sullerige
spraaklooze werkbeulerij, zonder klacht of krimp. En nou had ie
schik in zichzelf, dat ie daar weer stond, zijn Dirk, in gierende
windguurte, die door de kleeren heenrilde [3]op ’t naakte lijf, zonder
klacht. ’n Halven akker van Dirk stond Piet te spitten, gemakkelijker
grondbrok bewerkend, omdat ie niet had stille, wriklooze stierekracht
van Dirk. Vreemd zang-stemde, ver, heel ver, licht-bevend, dàn
opgejaagd door windrukken, zijn in deun-maat uitgezeurd lied,
dichtbij raar-vroolijk vergalmend bij elken zandworp, onder grauwe
luchtzwaarte. Vlak bij ouën Gerrit klankte nu door het winterveld,
zwak zanggestem in de bewogene lucht:— —Toe … en .… sprèk sai
van betooàle.… moar ’k hàt gain sint.…

Sai vloekte en skolde.… main f’r lailikke fint.… en wèg woei weer,
beverig-zwak Piet’s zeurig geluid in den wind over de
akkersombering vergàlmend.

—Hoho.… ho!.….. ho!.… die hep puur skik, grinnikte de Ouë,

afwerend met grimmige gezichtskreukels, aangutsend regengespat,
dat uit zware wolkdreiging over het land rafel-grauw heenvaalde
weer. Dirk hoorde niet in z’n spithei, spraakloos baggerde ie door in
z’n zandkuil, opwerpend rhytmisch, kleurige kluiten, blauwige zwarte
kleibonken, forsch-geweldig, in breede plakken, wegsnijend onder
z’n voeten.—

Met zware dubbele hypotheek belast, was het bunder land van den
ouën Gerrit, waarop z’n zoons nu werkten. Maar voor hem gelukkig,
lag ’t in het hartje van ’t tuinderstedeke Wiereland.—Z’n anderen
grond had ie gepacht, lag meer aan den kant van gemeente
Duinkijk, die komvormig om Wiereland liep en in één weg zich
uitstrekte recht-door naar zee, vormend daar één gemeente met
Zeekijk, armelijk zeegehucht, dat alleen zomers, door kleurige
vreugde van badgasten wat opleefde. Zijn bunder werd nou ’t laatst
bewerkt. Achter z’n huisje en erf liep ’t uit, omsloten rechts, door
groezelige armelijke huisachterbuurtjes uit de arbeiders- en
tuinderswijk, links openliggend, met aanschakeling van tuinders- en
bollenakkers, ver, heel ver.—Dooie tijd was ’t alom voor de tuinders
in ’t stedeke; luiertijd en tijd van ellende voor de meesten, pijnigende
zorg-armoe en grauw gebrek voor de Wierelandsche getrouwde
daggelders, los volk, dat met ’t sterven van den [4]zomer, werkeloos
rondscharrelde; hier ’n trap, daar ’n snauw, ’n paar dagen zwaar
spittend, dan weer niks, geradbraakt en omhongerend elken
winterdag, in gierig licht en zonnestraalte. De ouë Gerrit had z’n
twee pootige zoons, die hem te eten gaven, nou hij ’t jaren hun had
gedaan. Want nou, op z’n ouën dag, voelde ie zich na z’n zware
ziekte van verleden jaar te stram en te beroerd om in guurte en
regen te spitten. Als ie z’n kinderen niet had, zou ie nou nog op z’n
achtenzestigste jaar moèten, want loon kon ie niet betalen.

Lekker warmde ’t nou in ’m, dat hij maar met z’n handen in de
zakken daar te staan had, alleen maar kijken naar hun werk en ie z’n
eigen niet druk hoefde te maken. Wat ’n kerel was z’n Dirk toch!.…
hoho!.… wat ’n skoften.… wat ’n makelei.… wat ’n poote en
arreme!.… Jesis!.… wa stong ie daar stoer in z’n kiel.… huhu!.…
hu.… da skept moar, .… skiet moar òver. En s’n Piet gunters.… ook
’n pootige knoest.… kaik.… da song ie weer.… wat ’n hassebas.…
da se’m toch soo gemeen konne f’rbluffe.… kaik die Dirk.… da segt
t’met g’n woord.… stil.… da gong ie puur.… bònk!.… poef.… bònk!..
poef!! wat ’n vracht.… daa’s werk.… ieder keer t’ met vaiftig pond.…
boofe se macht.. main kristis.… Dâ ken puur nie eene van de
ploas!.…..

Ouë Gerrit huiver-kleumde van kou. De wind rumoer-baste en

kermloeide bolderig-wild over de vlakte, pal in zijn pimpelpaarse
gezichtskleum.… Jesis.… da natte gespikkel.… soo effe
sommetaids.… hai mos maor instappe.…

Dirk, in z’n spithei had ’n laatste kleiworp met hevige kracht-zwelling

en inkerving van z’n graaf, bevracht met heel z’n lijf-zwaarte, en
wierp uit, zwarter vastgestikte brokken, die blauw-gloederig nog
terugzogen in kwebbelenden moddergrond. Nou stond ie even uit te
hijgen, heet zweet van z’n brons gezwollen hoofd te luwen met vaal-
rooie zakdoek. En suffig, als timide waanzinnige, stomp, staarde ie
op het omwoelde grondbrok boven ’m, dat nu weer brons-kleurig lag
in het zwaar-grauwe druillicht, met z’n al ingedroogde kleur-plekken
aan de kanten, die geheimvol grijs, en teertonig tint-wasemden of
boetseer-bonkig en [5]en pulverig van rauwe rulheid woelden
tusschen roestige stapels, midden in zanderige uitgravingen. Met de
dikke vuilgroenige moddervingers van z’n geweldige, melaatsche
hand, gutste ie telkens ’t zweet van bronzen kop in spitkuil, en
telkens weer woei er nattig-splinterend, regen tegenaan. Nog even
uithijgen bleef-ie, weinig voelend van de bolderende windgieringen.
Dàt was werken, beestig zwaar, maar wat gaf ’t te donderjagen, ’t
mos toch dààn. Nou voelde ie z’n ruggestreng steken en pieken en
stram stroefde, lammig, ’n heete pijn in z’n polsen. Maar hij moest
ook nog wennen aan dien spit-buk-stand, uitputtend in zoo’n diepte.
Zoo dikwijls spitte ie ook niet vijf-steek. Als ie nou maar niet begon te
beven in z’n dondrement, want dan was ’t zaniken.… dan kon ie niet
meer voort. Net als verleden jaar, toen ie doorgong.… heere.… wat
beroerd was ie toen niet weest. Donders nou voelde ie de regen
weer, nattig geplak op z’n heet-dampend zweetlijf.… vastplakkerig.…
Dan maar weer aan den skep.

—Paás op Dirk! je hoalt natte beene.… pàas op, waarschuwde de

Ouë, die modderige opborreling van doorzuigend water zag
opschuimen uit den grond, rond Dirk’s klompen. Het schuimde onder
z’n beenen aan broek-onderend, nauw dichtgesnoerd met ontrafelde
brokken grauwen zak, als slobkousen slobberend. Op het
geschreeuw van den Ouë schrik-vlug, lichtte Dirk één vet-bekleid
been, met het andere even dieper vastzuigend in poelig zwart. Toen
veerkrachtig toch, versprong hij in andere hei-kuilholte op droog-
plek. Z’n klomp-voeten stonden vergrauwd plat-breed als
olifantsklauwen, rondig vereeld uit het kledderende kleizand. Nog
bleef ie uitstaren. Z’n stroo-gelig haar, grof-ongekamd, rossig
vervlekt, sluikte langs groote krom-ingekrulde ooren, en spriette
onder z’n pet, bij voorkop uit. Wit-blond vlasten z’n brauwen en kort,
als afgeknipt, zeerden wimpers om bloed-randige oogen, wat gaf z’n
beenigen chineezig ingekaakten kop iets versufts en wreeds. Z’n
dof-groene glazige oogen lobbesten maar stommig glansloos in z’n
grooten, wreeden kop. En aldoor keek ie voor-zich-uit, met iets van
koe-beestige loomheid in z’n doffe lenzen. [6]Spraakloos nog bleef ie
z’n gezicht bevegen, kalm door de krommige beenen van z’n vader
heenziend over de akkers. Zachter blaasbalgde z’n borst in lichten
kreun van ademhaling nà; zwaarste inspanning moest nog komen.

—Huhu!.… hu.… je laik puur warm, wà jou Dirk.… huhu!.…

huhu!.…, lachte de Ouë zuur, verlegen ’n beetje met eigen nietsdoen
en luierstand. Niets zei Dirk terug.… alleen in zachte uithijging
kermde gesmoord z’n borst en verstoomde adem. Even zou ie
omdraaien tegen wind in.…

Lager stond ie nu in z’n hei-kuil, en vóór ’m winterden en dorden de

uitgestorven akkers, bronzige wijdheid àlom, vergrauwd in knagende
droefnis. Van de verschrompelde erfjes en rottige tuinbrokjes achter
de huisjes uit Rozen- en Bikkerstraat die rechts Hassels land
begrensde, klapperde heftig geflapper van uitgespannen
waschgoed, boven-uit smerige schuttinkjes. Triestig en kniezerig
brokkeldakten de huisjes in de winter-sombering, met de oneffen
geveltjes die in lijnschokken hotsten tegen de laag-zware
luchtgrauwing. Vreemd-vaal kleurden òp, mos-groenig verweerd, en
omstoven met wintersmart de laag-gezonken droef-rooie
bedakingen. Heel de arbeiderswijk lag daar, met d’r gruizelig donker
klein gekrot als ingestort muurwerk te kijk; groenige zolderluikjes,
bruinige deurtjes, venster-knussige ruitjes, grijze gevel-stompjes,
peuterig-dorpsch, rommelig vergoord in de dagdonkering. De erf- en
tuintjes-schuttingen, uitrottend hout, kreunden in den gierwind. Van
Dirk’s hei af was te hooren het zeeruischend wild geslier en geloei,
van kaal-gewaaide iepen uit de achterafstraatjes, die takkennaakt
met forsch-donkre stamheffing, loei-zwiepend in hoog
overdonderend geweld, als reuzige wezens, boven de laag
knielende huisjes, hun winter-worsteling bevochten. De somberste
hoek lag te huiveren, vlak bij greppelrand en afwaterend slootje,
waar de achtergevel van ’n katholieke kerk, grijs-vaal, smartelijk
verkleurd in bruin-zwartig muurvocht hoog uitstak, met z’n weenende
stigma’s en kringen van verweerd kleurleed, rimpels van angst,
gekrast in gevelsteen; somberste hoek in de duistere druiling [7]en
kruiping van beangstigend wintergrauw. Daar, ver tegenover,
afgezien van erfjes-engte lag ’t uitgestorven land, waartusschen nog
niet omgespitte tuinderij, brokjes late andijvie, spruitjes en
koolhoekjes, wat armelijk geel-groen gevlek duwde. Maar overal,
wijd-om verschimmende bosschages, winterdroeve en kil-verre
hagen, afgeknaagd en vervreten, omprangd in één lucht-grauw,
laag, laag, met angstig zware voortschuiving van al donkerder
wolktinten, stug-droef nevelend en toch geheimvol vervalend over de
ontzettende, van-god-verlaten leege velden.

Juist wou Dirk z’n graaf weer in den grond steken, toen ie ouen Bolk
van het Wierelandsche achtereind, ’t droefste, afgekniesdste
ellendehoekje van de streek, naar hen toe zag dobberen.

—F’rjenne, dà hep je Jan Platneus, wà stapt dìe hier staif op an,.…

zei ie zacht, glurend tusschen vaders beenen door, die zich nog niet
had omgekeerd om te kijken.

’n Klein gedrongen mannetje strompelde waggelend-moeilijk op z’n

groote klompen tusschen de enge greppels van bollenakkers. Van
ver riep ie al, met doffe neusklankstem … gedaa’g, naar de Hassels.
Blaas-hijgend, in de handen slaand, kwam ie, klein-krom naast ouen